Skip to main content

Overview

Overview is the starting point: modeled total/reachable heap, object/class/array/root counts, size-model provenance, top retained objects, and Quick Checks.

Read the summary first​

Total heap includes modeled shallow bytes of recorded objects. Reachable heap uses the retaining graph. Neither equals file length or analyzer RSS. Size-model warnings describe uncertainty; hidden VM fields/class metadata are excluded.

Top retained objects and the heap-composition treemap/bar charts are selected ownership entries, not a disjoint partition. A collection, its array, and elements can describe overlapping storage.

Actions​

  • Select Why alive? on an object to inspect a root path.
  • Use Inspect / Show referrers where offered; IDs are scoped to the dump.
  • Copy Incident Report copies a local summary/report. It can include paths and previews; scrub before sharing.
  • Follow class/source actions into the project, then validate whether that owner is the repair location.

Quick Checks are heuristic leads. A high byte[] percentage identifies storage, not automatically a stream-close bug. Warnings/errors should be read as text as well as color.

Progress and recovery​

Early summary/progress updates can arrive before full analysis finishes; actions needing the completed graph may not yet be available. Hosts can present progress differently.

A dead per-tab server exposes Retry. Slow analysis offers waiting/cancellation rather than proving a crash. Closing one editor should not stop another editor's server. See troubleshooting.

Host availability distinguishes published VSIX targets from candidates. For investigation continue to Histogram.